Spatial modelling the location choice of large-scale solar photovoltaic power plants: Application of interpretable machine learning techniques and the national inventory

The optimum site selection of solar photovoltaics power plant across a given geographic space is usually assessed by using the geographic information system based multi-criteria decision making methods with various re-striction criteria, while such evaluation results vary with criteria weights and are difficult to be validated in real life practices. To address this issue, this paper uses a national inventory dataset of large-scale solar photovoltaics installations (the land coverage area & GE; 1 hm2) to investigate the spatial location choices of solar power plants with the aids of interpretable machine learning techniques. A total of 21 geospatial conditioning factors of solar energy development are considered. The location choices of solar photovoltaics installation are then modeled with the multi-Layer perceptron, random forest, extreme gradient boosting models for each land cover type (e.g. cropland, forest, grassland, and barren). The SHapley additive explanation and variable importance measure methods are adopted to identify key criteria and their influences on the solar photovoltaics installation location selection. Results indicate that the random forest model presented the better performance among three machine learning models. The relative importance of conditioning factors revealed that the vegetation index and distance to power grid were always the most important predictors of solar photovoltaics installation location. Further-more, topographical factors and transportation convenience may have a moderate impact on the spatial distri-bution of solar photovoltaics power stations. Unexpectedly, most of resources endowment and socio-economic factors play a negligible role in determining the optimal siting of solar power farms. Simulated solar photo-voltaics installations probability maps illustrated that the most suitable regions account for 4.6 % of China's total land area. The evidence-based method proposed in this research can not only help identify suitable solar pho-tovoltaics farm locations in terms of various decision-making criterion, but also provide a robust planning tool for sustainable development of solar energy sources.

Automated summary

In this study, interpretable machine learning techniques were used to investigate the spatial location choices of solar power plants based on a national inventory dataset of large-scale solar photovoltaics installations. The results showed that the random forest model performed the best, and vegetation index and distance to power grid were found to be the most important factors in the selection of solar photovoltaics installation locations.